Described herein are systems and methods for jointly learning geometry and motion with three-dimensional holistic understanding. In embodiments, such approaches enforce the inherent geometrical consistency during the learning process, yielding improved results for both tasks. In embodiments, three parallel networks are adopted to predict the camera motion (e.g., MotionNet), dense depth map (e.g., DepthNet), and per-pixel optical flow between consecutive frames (e.g., FlowNet), respectively. The information of 2D flow, camera pose, and depth maps, are fed into a holistic 3D motion parser (HMP) to disentangle and recover per-pixel 3D motion of both rigid background and moving objects. Various loss terms are formulated to jointly supervise the three networks. Embodiments of an efficient iterative training strategy are disclosed for better performance and more efficient convergence. Performance on depth estimation, optical flow estimation, odometry, moving object segmentation, and scene flow estimation demonstrates the effectiveness of the disclosed systems and methods.

Devices and a method are provided for providing feedback to a user. In one implementation, the method comprises obtaining a plurality of images from an image sensor. The image sensor is configured to be positioned for movement with the user's head. The method further comprises monitoring the images, and determining whether relative motion occurs between a first portion of a scene captured in the plurality of images and other portions of the scene captured in the plurality of images. If the first portion of the scene moves less than at least one other portion of the scene, the method comprises obtaining contextual information from the first portion of the scene. The method further comprises providing the feedback to the user based on at least part of the contextual information.

A defect detection method includes acquiring a reference image; selecting a target region of the reference image; identifying, based on a matching metric, one or more comparative regions of the reference image corresponding to the target region; acquiring a test image; masking the test image with the target region of the reference image and the one or more comparative regions of the reference image; defining a defect threshold for the target region in the test image based on the one or more comparative regions in the test image; and determining whether the target region of the test image contains a defect based on the defect threshold.

A system for controlling a vehicle includes a processor configured to execute instructions stored on a non-transitory computer readable medium. The system also includes a sensor coupled to the processor and configured to receive sensory input. The system also includes a controller coupled to the processor and configured to control the vehicle. The processor is further configured to: create a synthetic image based on the sensory input; derive a deep reinforcement learning (RL) policy using the synthetic image, wherein the deep RL policy determines a longitudinal control for the vehicle; and instruct the controller to control the vehicle based on the deep RL policy.

A mobile robot of the present disclosure includes a first pattern emission unit configured to emit a first patterned light downward and forward from the main body on a floor of an area to be cleaned; and an image acquisition unit configured to acquire an image of first patterned light emitted by the first pattern emission unit and incident on an obstacle. A pattern is detected from the acquired image to determine an obstacle, and a cliff is detected based on at least one of a shape or a position of the pattern in the image. The mobile robot may identify a travel path that does not lead to the cliff.

Systems and methods to analyze visible, near infrared, and thermal imagery of a land parcel in order to track and report changes to that land. A set of image processing techniques and algorithms can be utilized to generate composite images, align the imagery, compute differences, detect which features have changed, and meaningfully display the changed features. A moving window calculates differences in intensity between a first composite image and a second composite image and compares the differences against a threshold value. An interactive overlay user interface allows a user to adjust the magnitude and particular areas to be analyzed and presents the change.

Embodiments of the present disclosure pertain to extracting data corresponding to particular data types using neural networks. In one embodiment, a method includes receiving an image in a backend system, sending the image to an optical character recognition (OCR) component, and in accordance therewith, receiving a plurality of characters recognized in the image, sequentially processing the characters with a recurrent neural network to produce a plurality of outputs for each character, sequentially processing the plurality of outputs for each character with a masking neural network layer, and in accordance therewith, generating a first plurality of probabilities, wherein each probability corresponds to a particular character in the plurality of characters, selecting a second plurality of adjacent probabilities from the first plurality of probabilities that are above a threshold, and translating the second plurality of adjacent probabilities into output characters.

The invention provides the techniques and systems that allow for the identification and classification of objects (i.e., humans) in images using a predictive segmentation model. More specifically, human forms are identified within an image by generating pixel-level bounding boxes for each possible object and using offsets and segmentation masking. In some instances, embodiments of the invention use an identified floor plane that intersects with a bounding box to identify a three-dimensional position for the intersection point, which can then be assigned to the human form and represent its depth within the image.

Methods are provided for the automated detection and/or counting of micro-objects in a microfluidic device. In addition, methods are provided for repositioning micro-objects in a microfluidic device. In addition, methods are provided for separating micro-objects in a spatial region of the microfluidic device.

Proceeds deposit machine has determination setting data containing correspondence between types of the coupons and information relating to necessity of authentication determination. A type determination unit determines the type of deposited coupon and then refers to the determination setting data. If yes has been set for the necessity of the authentication determination for the determined type of coupon, an authentication determination unit performs authentication determination. If the coupon is determined to be counterfeit as result of the authentication determination, the proceeds deposit machine discharges this coupon. If the coupon is determined to be genuine, proceeds deposit machine stores this coupon in one of sheet cassettes. If no has been set for the necessity of the authentication determination for the determined type of the coupon, proceeds deposit machine stores this coupon in one of the sheet cassettes without performing the authentication determination by the authentication determination unit.